One import.
Enterprise infrastructure.
Conduit is the production-grade MCP client for DataGrout Hub. Swap one import and your agent gets mTLS identity, semantic tool discovery, 1,000+ enterprise tools, and governed execution โ in Python, TypeScript, Rust, Elixir, or Ruby.
The before and after.
Every agent project starts with the same infrastructure tax. Conduit eliminates it.
# Before: every agent project starts with this tax
import requests, time, json
from salesforce_bulk import SalesforceBulk
# Auth setup โ 30+ lines
bulk = SalesforceBulk(
username=os.environ["SF_USER"],
password=os.environ["SF_PASS"],
security_token=os.environ["SF_TOKEN"]
)
# Retry logic you have to write yourself
def fetch_with_retry(job, max_retries=3):
for attempt in range(max_retries):
try:
return bulk.get_all_results_for_query(job)
except Exception:
if attempt == max_retries - 1: raise
time.sleep(2 ** attempt)
# ...and you haven't written a single line of agent logic yet.# After: one import, everything else is handled
from datagrout.conduit import Client
async with Client(
"https://gateway.datagrout.ai/servers/{uuid}/mcp",
auth={"bearer": "your-access-token"}
# mTLS: call Client.bootstrap_identity() once โ no tokens needed afterward
) as client:
# Find tools by natural language โ not by memorizing API names
results = await client.discover(
query="get overdue invoices from last 30 days",
limit=5
)
# Execute โ auth, pagination, retries all handled
result = await client.perform(results[0].tool.name, {})What you didn't have to write:
Everything that comes with the import.
Conduit isn't just a connection helper. It's the full production-grade infrastructure layer for your agent.
mTLS identity, one-time bootstrap
Call bootstrap_identity() once โ the SDK generates a keypair, registers with the DataGrout CA, and saves certs locally. All subsequent runs auto-discover and renew automatically. No manual PKI setup.
1,000+ enterprise tools, zero auth code
Salesforce (720+ tools), QuickBooks (560+ tools), and more โ OAuth2, pagination, and retry logic already handled.
Semantic discovery, not name-guessing
Pass a natural-language goal, get back the ranked tools that match. No memorizing API surface areas.
Composes with Logic for persistent memory
client.logic gives your agents access to persistent symbolic facts that survive between calls โ shared across all agents on your server. Conduit wires the connection; Logic owns the memory.
Governance baked in
Budget caps, policy enforcement, prompt injection defense, and loop detection โ without writing any of it yourself.
cache_ref chaining
Pipeline data server-side across tool calls. Your agent only sees the final result โ no re-fetching, no token overhead.
Namespaced tool wrappers
client.prism, client.logic, client.warden, client.flow, client.ephemerals โ idiomatic, typed wrappers for every DataGrout tool surface. No string-matching tool names.
Batch operations
perform_batch sends multiple tool calls in a single request. Fewer round-trips, lower latency, same governance and receipts on every call.
Cost tracking metadata
Every response includes credits consumed, model used, and token counts โ accessible via _meta.datagrout. Three-tier fallback format so no response ever drops the receipt.
Start in 60 seconds.
Install the SDK, connect to Hub, and discover your first tool โ all before your coffee gets cold.
MIT licensed ยท Apache 2.0 compatible
pip install datagrout-conduit==0.3.0
from datagrout.conduit import Client
async with Client(
"https://gateway.datagrout.ai/servers/{uuid}/mcp",
auth={"bearer": "your-access-token"}
) as client:
tools = await client.list_tools()results = await client.discover(query="your goal here", limit=5)
result = await client.perform(results[0].tool.name, {})Your agent. Enterprise infrastructure. One import.
Connect to Hub, discover tools by goal, and ship โ without writing auth, pagination, or retry glue.
